An empirical approach to modelling climate change impact on building energy use

This paper aims to demonstrate an empirical approach to building energy modelling (BEM). The study presented here uses machine learning methods with metered data in substitute to detailed on site surveys in conventional BEM simulations. Key advantages of the machine learning approach are the reduced site survey time and having a site calibrated model, giving a much shorter lead in and pre-calibrated model for the site. The increased availability of metered data in modern buildings makes this type of approach more viable than in the past. In demonstrating this methodology, this paper will choose to predict electrical cooling energy use for the Information Commons building at the University of Sheffield, in response to projected future and past weather using Gaussian processes.